Project Summary/Abstract TR&D 2 Statistical Inference for Imaging and Disease We aim to close the gap between the wealth of information captured by brain MRI scans and the simple neuroimage-derived phenotypes, such as volume of white matter hyperintensity, used in neuroimaging studies. We will develop machine learning methods to enable detailed quantitative characterization of anatomical patterns of disease from images routinely acquired in clinical practice. The proposed methods will bring the vast collections of clinically acquired brain MRI scans to bear on important medical problems. The analysis will produce rich hypotheses for research studies based on detailed phenotypes of anatomical patterns of disease, enable image-based patient assessment and improve disease monitoring. The proposed approach is based on two key innovations. First, we will develop statistical inference methods to enable the application of state-of-the-art medical image computing algorithms, originally designed for high resolution isotropic brain MRI, to clinical images with sparse slice sampling. Second, we will develop machine learning methods for creating detailed phenotypes to characterize spatial patterns of white matter disease. We will develop the proposed techniques in the context of clinically important studies of white matter disease and its influence on stroke recovery. We hypothesize that the novel spatial phenotypes extracted from clinical brain MRI will improve prediction of stroke outcomes and produce refined mechanistic hypotheses about the underlying disease processes. We will validate and deploy the proposed methods in close collaboration with the Imaging Core of the International Stroke Genetics Consortium. We will pursue methodological synergies with the other TR&D projects and will work to advance knowledge and to improve patient assessment in a broad range of clinical problems defined by the Collaboration Projects and Service Projects.